Online ARIMA

hanaml.OnlineARIMA is a R wrapper

hanaml.OnlineARIMA(
  order = NULL,
  learning.rate = NULL,
  epsilon = NULL,
  output.fitted = TRUE,
  random.state = NULL,
  random.initialization = NULL
)

Arguments

order

list/vector of integer, optional
Indicate the order (p, d, m).

p: value of the auto regression order.
d: value of the differentiation order.
m: extended order needed to transform all moving-averaging term to AR term.

Defaults to (1, 0, 0).

learning.rate

double, optional
Learning rate. Must be greater than zero.
Default value is calculated according to order(p, d, m).

epsilon

double, optional
Convergence criterion.
Default value is calculated according to learning.rate, p and m in the order.

output.fitted

logical, optional
Output fitted result and residuals if TRUE.
Defaults to TRUE.

random.state

integer, optional
Specifies the seed for random number generator.

0: use the current time (in second) as seed.
Others: use the specified value as seed.

Default to 0.

random.initialization

logical, optional
Whether randomly generate initial state.

FALSE: set all to zero.
TRUE: use random values.

Defaults to FALSE.

Value

Returns an "OnlineARIMA" object and then invoke the fit function will obtain the following attributes:

model: DataFrame
Fitted model.
fitted: DataFrame
Predicted dependent variable values for training data. Set to NULL if the training data has no row IDs.

Details

Online ARIMA implements an online learning method to estimate the parameters of ARIMA models by reformulating it into a full information online optimization task (without random noise terms), which has no limitations of depending on noise terms and accessing the entire large-scale dataset in advance.

Methods

fit(data = NULL, key = NULL, endog = NULL, learning.rate = NULL, epsilon = NULL, ...) The fit function of OnlineARIMA object. Usage:
onlinearima <- hanaml.OnlineARIMA(order=c(4,0,8), output.fitted=TRUE, learning.rate=0.00001)
onlinearima$fit(data) Arguments: data, DataFrame
Input data. key, character, optional
The timestamp column of data.
The type of key column is int.
Defaults to the first column of data if not provided. endog, character, optional
The endogenous variable, i.e. time series.
Defaults to the first non-key column of data if not provided. learning.rate, double, optional
Learning rate. Must be greater than zero.
Default value is calculated according to order(p, d, m). epsilon, double, optional
Convergence criterion.
Default value is calculated according to learning.rate, p and m in the order. ...
Reserved parameter.

Examples

Input DataFrame data:


> data$Collect()
    TIMESTAMP     Y
1           1   450
2           2   806
3           3   647
4           4 -3300
5           5  5308
......

Create a OnlineARIMA object without the data:


> onlinearima <- hanaml.OnlineARIMA(order=c(4,0,8), output.fitted=TRUE, learning.rate=0.00001)

Invoke fit function:


> onlinearima$fit(data)

Output:


> onlinearima$model
      KEY        VALUE
1   lrate        1e-05
2  fitted            1
3       d            0
4      mp           12
5 epsilon      2.4e-09
 .......
> onlinearima$fitted
    TIMESTAMP        FITTED    RESIDUALS
1           1    900.000000   -450.00000
2           2   1612.000000   -806.00000
3           3   1182.976801   -535.97680
4           4  -6855.679157   3555.67916
......

New coming input DataFrame data2:


> data2$Collect()
     TIMESTAMP      Y
1         101    386
2         102  -7807
3         103   3374
4         104   6074
5         105    241
......

Invoke the fit function again, learning.rate and epsilon could be re-assigned:

onlinearima$fit(data2)